1 /*
   2  * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "interp_masm_sparc.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "oops/arrayOop.hpp"
  30 #include "oops/markOop.hpp"
  31 #include "oops/methodData.hpp"
  32 #include "oops/method.hpp"
  33 #include "oops/methodCounters.hpp"
  34 #include "prims/jvmtiExport.hpp"
  35 #include "prims/jvmtiRedefineClassesTrace.hpp"
  36 #include "prims/jvmtiThreadState.hpp"
  37 #include "runtime/basicLock.hpp"
  38 #include "runtime/biasedLocking.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "runtime/thread.inline.hpp"
  41 
  42 #ifndef CC_INTERP
  43 #ifndef FAST_DISPATCH
  44 #define FAST_DISPATCH 1
  45 #endif
  46 #undef FAST_DISPATCH
  47 
  48 // Implementation of InterpreterMacroAssembler
  49 
  50 // This file specializes the assember with interpreter-specific macros
  51 
  52 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS);
  53 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS);
  54 
  55 #else // CC_INTERP
  56 #ifndef STATE
  57 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name))
  58 #endif // STATE
  59 
  60 #endif // CC_INTERP
  61 
  62 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) {
  63   // Note: this algorithm is also used by C1's OSR entry sequence.
  64   // Any changes should also be applied to CodeEmitter::emit_osr_entry().
  65   assert_different_registers(args_size, locals_size);
  66   // max_locals*2 for TAGS.  Assumes that args_size has already been adjusted.
  67   subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words
  68   // Use br/mov combination because it works on both V8 and V9 and is
  69   // faster.
  70   Label skip_move;
  71   br(Assembler::negative, true, Assembler::pt, skip_move);
  72   delayed()->mov(G0, delta);
  73   bind(skip_move);
  74   round_to(delta, WordsPerLong);       // make multiple of 2 (SP must be 2-word aligned)
  75   sll(delta, LogBytesPerWord, delta);  // extra space for locals in bytes
  76 }
  77 
  78 #ifndef CC_INTERP
  79 
  80 // Dispatch code executed in the prolog of a bytecode which does not do it's
  81 // own dispatch. The dispatch address is computed and placed in IdispatchAddress
  82 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
  83   assert_not_delayed();
  84 #ifdef FAST_DISPATCH
  85   // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
  86   // they both use I2.
  87   assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
  88   ldub(Lbcp, bcp_incr, Lbyte_code);                     // load next bytecode
  89   add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
  90                                                         // add offset to correct dispatch table
  91   sll(Lbyte_code, LogBytesPerWord, Lbyte_code);         // multiply by wordSize
  92   ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr
  93 #else
  94   ldub( Lbcp, bcp_incr, Lbyte_code);                    // load next bytecode
  95   // dispatch table to use
  96   AddressLiteral tbl(Interpreter::dispatch_table(state));
  97   sll(Lbyte_code, LogBytesPerWord, Lbyte_code);         // multiply by wordSize
  98   set(tbl, G3_scratch);                                 // compute addr of table
  99   ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress);     // get entry addr
 100 #endif
 101 }
 102 
 103 
 104 // Dispatch code executed in the epilog of a bytecode which does not do it's
 105 // own dispatch. The dispatch address in IdispatchAddress is used for the
 106 // dispatch.
 107 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
 108   assert_not_delayed();
 109   verify_FPU(1, state);
 110   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 111   jmp( IdispatchAddress, 0 );
 112   if (bcp_incr != 0)  delayed()->inc(Lbcp, bcp_incr);
 113   else                delayed()->nop();
 114 }
 115 
 116 
 117 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) {
 118   // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
 119   assert_not_delayed();
 120   ldub( Lbcp, bcp_incr, Lbyte_code);               // load next bytecode
 121   dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr);
 122 }
 123 
 124 
 125 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) {
 126   // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
 127   assert_not_delayed();
 128   ldub( Lbcp, bcp_incr, Lbyte_code);               // load next bytecode
 129   dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false);
 130 }
 131 
 132 
 133 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 134   // load current bytecode
 135   assert_not_delayed();
 136   ldub( Lbcp, 0, Lbyte_code);               // load next bytecode
 137   dispatch_base(state, table);
 138 }
 139 
 140 
 141 void InterpreterMacroAssembler::call_VM_leaf_base(
 142   Register java_thread,
 143   address  entry_point,
 144   int      number_of_arguments
 145 ) {
 146   if (!java_thread->is_valid())
 147     java_thread = L7_thread_cache;
 148   // super call
 149   MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments);
 150 }
 151 
 152 
 153 void InterpreterMacroAssembler::call_VM_base(
 154   Register        oop_result,
 155   Register        java_thread,
 156   Register        last_java_sp,
 157   address         entry_point,
 158   int             number_of_arguments,
 159   bool            check_exception
 160 ) {
 161   if (!java_thread->is_valid())
 162     java_thread = L7_thread_cache;
 163   // See class ThreadInVMfromInterpreter, which assumes that the interpreter
 164   // takes responsibility for setting its own thread-state on call-out.
 165   // However, ThreadInVMfromInterpreter resets the state to "in_Java".
 166 
 167   //save_bcp();                                  // save bcp
 168   MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception);
 169   //restore_bcp();                               // restore bcp
 170   //restore_locals();                            // restore locals pointer
 171 }
 172 
 173 
 174 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
 175   if (JvmtiExport::can_pop_frame()) {
 176     Label L;
 177 
 178     // Check the "pending popframe condition" flag in the current thread
 179     ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg);
 180 
 181     // Initiate popframe handling only if it is not already being processed.  If the flag
 182     // has the popframe_processing bit set, it means that this code is called *during* popframe
 183     // handling - we don't want to reenter.
 184     btst(JavaThread::popframe_pending_bit, scratch_reg);
 185     br(zero, false, pt, L);
 186     delayed()->nop();
 187     btst(JavaThread::popframe_processing_bit, scratch_reg);
 188     br(notZero, false, pt, L);
 189     delayed()->nop();
 190 
 191     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 192     // address of the same-named entrypoint in the generated interpreter code.
 193     call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 194 
 195     // Jump to Interpreter::_remove_activation_preserving_args_entry
 196     jmpl(O0, G0, G0);
 197     delayed()->nop();
 198     bind(L);
 199   }
 200 }
 201 
 202 
 203 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 204   Register thr_state = G4_scratch;
 205   ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
 206   const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset());
 207   const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset());
 208   const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset());
 209   switch (state) {
 210   case ltos: ld_long(val_addr, Otos_l);                   break;
 211   case atos: ld_ptr(oop_addr, Otos_l);
 212              st_ptr(G0, oop_addr);                        break;
 213   case btos:                                           // fall through
 214   case ctos:                                           // fall through
 215   case stos:                                           // fall through
 216   case itos: ld(val_addr, Otos_l1);                       break;
 217   case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break;
 218   case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break;
 219   case vtos: /* nothing to do */                          break;
 220   default  : ShouldNotReachHere();
 221   }
 222   // Clean up tos value in the jvmti thread state
 223   or3(G0, ilgl, G3_scratch);
 224   stw(G3_scratch, tos_addr);
 225   st_long(G0, val_addr);
 226   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 227 }
 228 
 229 
 230 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
 231   if (JvmtiExport::can_force_early_return()) {
 232     Label L;
 233     Register thr_state = G3_scratch;
 234     ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
 235     br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit;
 236 
 237     // Initiate earlyret handling only if it is not already being processed.
 238     // If the flag has the earlyret_processing bit set, it means that this code
 239     // is called *during* earlyret handling - we don't want to reenter.
 240     ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch);
 241     cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L);
 242 
 243     // Call Interpreter::remove_activation_early_entry() to get the address of the
 244     // same-named entrypoint in the generated interpreter code
 245     ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1);
 246     call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1);
 247 
 248     // Jump to Interpreter::_remove_activation_early_entry
 249     jmpl(O0, G0, G0);
 250     delayed()->nop();
 251     bind(L);
 252   }
 253 }
 254 
 255 
 256 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) {
 257   mov(arg_1, O0);
 258   mov(arg_2, O1);
 259   MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2);
 260 }
 261 #endif /* CC_INTERP */
 262 
 263 
 264 #ifndef CC_INTERP
 265 
 266 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) {
 267   assert_not_delayed();
 268   dispatch_Lbyte_code(state, table);
 269 }
 270 
 271 
 272 void InterpreterMacroAssembler::dispatch_normal(TosState state) {
 273   dispatch_base(state, Interpreter::normal_table(state));
 274 }
 275 
 276 
 277 void InterpreterMacroAssembler::dispatch_only(TosState state) {
 278   dispatch_base(state, Interpreter::dispatch_table(state));
 279 }
 280 
 281 
 282 // common code to dispatch and dispatch_only
 283 // dispatch value in Lbyte_code and increment Lbcp
 284 
 285 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) {
 286   verify_FPU(1, state);
 287   // %%%%% maybe implement +VerifyActivationFrameSize here
 288   //verify_thread(); //too slow; we will just verify on method entry & exit
 289   if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 290 #ifdef FAST_DISPATCH
 291   if (table == Interpreter::dispatch_table(state)) {
 292     // use IdispatchTables
 293     add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
 294                                                         // add offset to correct dispatch table
 295     sll(Lbyte_code, LogBytesPerWord, Lbyte_code);       // multiply by wordSize
 296     ld_ptr(IdispatchTables, Lbyte_code, G3_scratch);    // get entry addr
 297   } else {
 298 #endif
 299     // dispatch table to use
 300     AddressLiteral tbl(table);
 301     sll(Lbyte_code, LogBytesPerWord, Lbyte_code);       // multiply by wordSize
 302     set(tbl, G3_scratch);                               // compute addr of table
 303     ld_ptr(G3_scratch, Lbyte_code, G3_scratch);         // get entry addr
 304 #ifdef FAST_DISPATCH
 305   }
 306 #endif
 307   jmp( G3_scratch, 0 );
 308   if (bcp_incr != 0)  delayed()->inc(Lbcp, bcp_incr);
 309   else                delayed()->nop();
 310 }
 311 
 312 
 313 // Helpers for expression stack
 314 
 315 // Longs and doubles are Category 2 computational types in the
 316 // JVM specification (section 3.11.1) and take 2 expression stack or
 317 // local slots.
 318 // Aligning them on 32 bit with tagged stacks is hard because the code generated
 319 // for the dup* bytecodes depends on what types are already on the stack.
 320 // If the types are split into the two stack/local slots, that is much easier
 321 // (and we can use 0 for non-reference tags).
 322 
 323 // Known good alignment in _LP64 but unknown otherwise
 324 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) {
 325   assert_not_delayed();
 326 
 327 #ifdef _LP64
 328   ldf(FloatRegisterImpl::D, r1, offset, d);
 329 #else
 330   ldf(FloatRegisterImpl::S, r1, offset, d);
 331   ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor());
 332 #endif
 333 }
 334 
 335 // Known good alignment in _LP64 but unknown otherwise
 336 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) {
 337   assert_not_delayed();
 338 
 339 #ifdef _LP64
 340   stf(FloatRegisterImpl::D, d, r1, offset);
 341   // store something more useful here
 342   debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
 343 #else
 344   stf(FloatRegisterImpl::S, d, r1, offset);
 345   stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize);
 346 #endif
 347 }
 348 
 349 
 350 // Known good alignment in _LP64 but unknown otherwise
 351 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) {
 352   assert_not_delayed();
 353 #ifdef _LP64
 354   ldx(r1, offset, rd);
 355 #else
 356   ld(r1, offset, rd);
 357   ld(r1, offset + Interpreter::stackElementSize, rd->successor());
 358 #endif
 359 }
 360 
 361 // Known good alignment in _LP64 but unknown otherwise
 362 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) {
 363   assert_not_delayed();
 364 
 365 #ifdef _LP64
 366   stx(l, r1, offset);
 367   // store something more useful here
 368   debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
 369 #else
 370   st(l, r1, offset);
 371   st(l->successor(), r1, offset + Interpreter::stackElementSize);
 372 #endif
 373 }
 374 
 375 void InterpreterMacroAssembler::pop_i(Register r) {
 376   assert_not_delayed();
 377   ld(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 378   inc(Lesp, Interpreter::stackElementSize);
 379   debug_only(verify_esp(Lesp));
 380 }
 381 
 382 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) {
 383   assert_not_delayed();
 384   ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 385   inc(Lesp, Interpreter::stackElementSize);
 386   debug_only(verify_esp(Lesp));
 387 }
 388 
 389 void InterpreterMacroAssembler::pop_l(Register r) {
 390   assert_not_delayed();
 391   load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 392   inc(Lesp, 2*Interpreter::stackElementSize);
 393   debug_only(verify_esp(Lesp));
 394 }
 395 
 396 
 397 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) {
 398   assert_not_delayed();
 399   ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f);
 400   inc(Lesp, Interpreter::stackElementSize);
 401   debug_only(verify_esp(Lesp));
 402 }
 403 
 404 
 405 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) {
 406   assert_not_delayed();
 407   load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f);
 408   inc(Lesp, 2*Interpreter::stackElementSize);
 409   debug_only(verify_esp(Lesp));
 410 }
 411 
 412 
 413 void InterpreterMacroAssembler::push_i(Register r) {
 414   assert_not_delayed();
 415   debug_only(verify_esp(Lesp));
 416   st(r, Lesp, 0);
 417   dec(Lesp, Interpreter::stackElementSize);
 418 }
 419 
 420 void InterpreterMacroAssembler::push_ptr(Register r) {
 421   assert_not_delayed();
 422   st_ptr(r, Lesp, 0);
 423   dec(Lesp, Interpreter::stackElementSize);
 424 }
 425 
 426 // remember: our convention for longs in SPARC is:
 427 // O0 (Otos_l1) has high-order part in first word,
 428 // O1 (Otos_l2) has low-order part in second word
 429 
 430 void InterpreterMacroAssembler::push_l(Register r) {
 431   assert_not_delayed();
 432   debug_only(verify_esp(Lesp));
 433   // Longs are stored in memory-correct order, even if unaligned.
 434   int offset = -Interpreter::stackElementSize;
 435   store_unaligned_long(r, Lesp, offset);
 436   dec(Lesp, 2 * Interpreter::stackElementSize);
 437 }
 438 
 439 
 440 void InterpreterMacroAssembler::push_f(FloatRegister f) {
 441   assert_not_delayed();
 442   debug_only(verify_esp(Lesp));
 443   stf(FloatRegisterImpl::S, f, Lesp, 0);
 444   dec(Lesp, Interpreter::stackElementSize);
 445 }
 446 
 447 
 448 void InterpreterMacroAssembler::push_d(FloatRegister d)   {
 449   assert_not_delayed();
 450   debug_only(verify_esp(Lesp));
 451   // Longs are stored in memory-correct order, even if unaligned.
 452   int offset = -Interpreter::stackElementSize;
 453   store_unaligned_double(d, Lesp, offset);
 454   dec(Lesp, 2 * Interpreter::stackElementSize);
 455 }
 456 
 457 
 458 void InterpreterMacroAssembler::push(TosState state) {
 459   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 460   switch (state) {
 461     case atos: push_ptr();            break;
 462     case btos: push_i();              break;
 463     case ctos:
 464     case stos: push_i();              break;
 465     case itos: push_i();              break;
 466     case ltos: push_l();              break;
 467     case ftos: push_f();              break;
 468     case dtos: push_d();              break;
 469     case vtos: /* nothing to do */    break;
 470     default  : ShouldNotReachHere();
 471   }
 472 }
 473 
 474 
 475 void InterpreterMacroAssembler::pop(TosState state) {
 476   switch (state) {
 477     case atos: pop_ptr();            break;
 478     case btos: pop_i();              break;
 479     case ctos:
 480     case stos: pop_i();              break;
 481     case itos: pop_i();              break;
 482     case ltos: pop_l();              break;
 483     case ftos: pop_f();              break;
 484     case dtos: pop_d();              break;
 485     case vtos: /* nothing to do */   break;
 486     default  : ShouldNotReachHere();
 487   }
 488   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 489 }
 490 
 491 
 492 // Helpers for swap and dup
 493 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 494   ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val);
 495 }
 496 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 497   st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n));
 498 }
 499 
 500 
 501 void InterpreterMacroAssembler::load_receiver(Register param_count,
 502                                               Register recv) {
 503   sll(param_count, Interpreter::logStackElementSize, param_count);
 504   ld_ptr(Lesp, param_count, recv);  // gets receiver oop
 505 }
 506 
 507 void InterpreterMacroAssembler::empty_expression_stack() {
 508   // Reset Lesp.
 509   sub( Lmonitors, wordSize, Lesp );
 510 
 511   // Reset SP by subtracting more space from Lesp.
 512   Label done;
 513   assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!");
 514 
 515   // A native does not need to do this, since its callee does not change SP.
 516   ld(Lmethod, Method::access_flags_offset(), Gframe_size);  // Load access flags.
 517   btst(JVM_ACC_NATIVE, Gframe_size);
 518   br(Assembler::notZero, false, Assembler::pt, done);
 519   delayed()->nop();
 520 
 521   // Compute max expression stack+register save area
 522   ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size);
 523   lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size);  // Load max stack.
 524   add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size );
 525 
 526   //
 527   // now set up a stack frame with the size computed above
 528   //
 529   //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below
 530   sll( Gframe_size, LogBytesPerWord, Gframe_size );
 531   sub( Lesp, Gframe_size, Gframe_size );
 532   and3( Gframe_size, -(2 * wordSize), Gframe_size );          // align SP (downwards) to an 8/16-byte boundary
 533   debug_only(verify_sp(Gframe_size, G4_scratch));
 534 #ifdef _LP64
 535   sub(Gframe_size, STACK_BIAS, Gframe_size );
 536 #endif
 537   mov(Gframe_size, SP);
 538 
 539   bind(done);
 540 }
 541 
 542 
 543 #ifdef ASSERT
 544 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) {
 545   Label Bad, OK;
 546 
 547   // Saved SP must be aligned.
 548 #ifdef _LP64
 549   btst(2*BytesPerWord-1, Rsp);
 550 #else
 551   btst(LongAlignmentMask, Rsp);
 552 #endif
 553   br(Assembler::notZero, false, Assembler::pn, Bad);
 554   delayed()->nop();
 555 
 556   // Saved SP, plus register window size, must not be above FP.
 557   add(Rsp, frame::register_save_words * wordSize, Rtemp);
 558 #ifdef _LP64
 559   sub(Rtemp, STACK_BIAS, Rtemp);  // Bias Rtemp before cmp to FP
 560 #endif
 561   cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad);
 562 
 563   // Saved SP must not be ridiculously below current SP.
 564   size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K);
 565   set(maxstack, Rtemp);
 566   sub(SP, Rtemp, Rtemp);
 567 #ifdef _LP64
 568   add(Rtemp, STACK_BIAS, Rtemp);  // Unbias Rtemp before cmp to Rsp
 569 #endif
 570   cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad);
 571 
 572   ba_short(OK);
 573 
 574   bind(Bad);
 575   stop("on return to interpreted call, restored SP is corrupted");
 576 
 577   bind(OK);
 578 }
 579 
 580 
 581 void InterpreterMacroAssembler::verify_esp(Register Resp) {
 582   // about to read or write Resp[0]
 583   // make sure it is not in the monitors or the register save area
 584   Label OK1, OK2;
 585 
 586   cmp(Resp, Lmonitors);
 587   brx(Assembler::lessUnsigned, true, Assembler::pt, OK1);
 588   delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp);
 589   stop("too many pops:  Lesp points into monitor area");
 590   bind(OK1);
 591 #ifdef _LP64
 592   sub(Resp, STACK_BIAS, Resp);
 593 #endif
 594   cmp(Resp, SP);
 595   brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2);
 596   delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp);
 597   stop("too many pushes:  Lesp points into register window");
 598   bind(OK2);
 599 }
 600 #endif // ASSERT
 601 
 602 // Load compiled (i2c) or interpreter entry when calling from interpreted and
 603 // do the call. Centralized so that all interpreter calls will do the same actions.
 604 // If jvmti single stepping is on for a thread we must not call compiled code.
 605 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) {
 606 
 607   // Assume we want to go compiled if available
 608 
 609   ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target);
 610 
 611   if (JvmtiExport::can_post_interpreter_events()) {
 612     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 613     // compiled code in threads for which the event is enabled.  Check here for
 614     // interp_only_mode if these events CAN be enabled.
 615     verify_thread();
 616     Label skip_compiled_code;
 617 
 618     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
 619     ld(interp_only, scratch);
 620     cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn);
 621     delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target);
 622     bind(skip_compiled_code);
 623   }
 624 
 625   // the i2c_adapters need Method* in G5_method (right? %%%)
 626   // do the call
 627 #ifdef ASSERT
 628   {
 629     Label ok;
 630     br_notnull_short(target, Assembler::pt, ok);
 631     stop("null entry point");
 632     bind(ok);
 633   }
 634 #endif // ASSERT
 635 
 636   // Adjust Rret first so Llast_SP can be same as Rret
 637   add(Rret, -frame::pc_return_offset, O7);
 638   add(Lesp, BytesPerWord, Gargs); // setup parameter pointer
 639   // Record SP so we can remove any stack space allocated by adapter transition
 640   jmp(target, 0);
 641   delayed()->mov(SP, Llast_SP);
 642 }
 643 
 644 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) {
 645   assert_not_delayed();
 646 
 647   Label not_taken;
 648   if (ptr_compare) brx(cc, false, Assembler::pn, not_taken);
 649   else             br (cc, false, Assembler::pn, not_taken);
 650   delayed()->nop();
 651 
 652   TemplateTable::branch(false,false);
 653 
 654   bind(not_taken);
 655 
 656   profile_not_taken_branch(G3_scratch);
 657 }
 658 
 659 
 660 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(
 661                                   int         bcp_offset,
 662                                   Register    Rtmp,
 663                                   Register    Rdst,
 664                                   signedOrNot is_signed,
 665                                   setCCOrNot  should_set_CC ) {
 666   assert(Rtmp != Rdst, "need separate temp register");
 667   assert_not_delayed();
 668   switch (is_signed) {
 669    default: ShouldNotReachHere();
 670 
 671    case   Signed:  ldsb( Lbcp, bcp_offset, Rdst  );  break; // high byte
 672    case Unsigned:  ldub( Lbcp, bcp_offset, Rdst  );  break; // high byte
 673   }
 674   ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte
 675   sll( Rdst, BitsPerByte, Rdst);
 676   switch (should_set_CC ) {
 677    default: ShouldNotReachHere();
 678 
 679    case      set_CC:  orcc( Rdst, Rtmp, Rdst ); break;
 680    case dont_set_CC:  or3(  Rdst, Rtmp, Rdst ); break;
 681   }
 682 }
 683 
 684 
 685 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(
 686                                   int        bcp_offset,
 687                                   Register   Rtmp,
 688                                   Register   Rdst,
 689                                   setCCOrNot should_set_CC ) {
 690   assert(Rtmp != Rdst, "need separate temp register");
 691   assert_not_delayed();
 692   add( Lbcp, bcp_offset, Rtmp);
 693   andcc( Rtmp, 3, G0);
 694   Label aligned;
 695   switch (should_set_CC ) {
 696    default: ShouldNotReachHere();
 697 
 698    case      set_CC: break;
 699    case dont_set_CC: break;
 700   }
 701 
 702   br(Assembler::zero, true, Assembler::pn, aligned);
 703 #ifdef _LP64
 704   delayed()->ldsw(Rtmp, 0, Rdst);
 705 #else
 706   delayed()->ld(Rtmp, 0, Rdst);
 707 #endif
 708 
 709   ldub(Lbcp, bcp_offset + 3, Rdst);
 710   ldub(Lbcp, bcp_offset + 2, Rtmp);  sll(Rtmp,  8, Rtmp);  or3(Rtmp, Rdst, Rdst);
 711   ldub(Lbcp, bcp_offset + 1, Rtmp);  sll(Rtmp, 16, Rtmp);  or3(Rtmp, Rdst, Rdst);
 712 #ifdef _LP64
 713   ldsb(Lbcp, bcp_offset + 0, Rtmp);  sll(Rtmp, 24, Rtmp);
 714 #else
 715   // Unsigned load is faster than signed on some implementations
 716   ldub(Lbcp, bcp_offset + 0, Rtmp);  sll(Rtmp, 24, Rtmp);
 717 #endif
 718   or3(Rtmp, Rdst, Rdst );
 719 
 720   bind(aligned);
 721   if (should_set_CC == set_CC) tst(Rdst);
 722 }
 723 
 724 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index,
 725                                                        int bcp_offset, size_t index_size) {
 726   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 727   if (index_size == sizeof(u2)) {
 728     get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned);
 729   } else if (index_size == sizeof(u4)) {
 730     assert(EnableInvokeDynamic, "giant index used only for JSR 292");
 731     get_4_byte_integer_at_bcp(bcp_offset, temp, index);
 732     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 733     xor3(index, -1, index);  // convert to plain index
 734   } else if (index_size == sizeof(u1)) {
 735     ldub(Lbcp, bcp_offset, index);
 736   } else {
 737     ShouldNotReachHere();
 738   }
 739 }
 740 
 741 
 742 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp,
 743                                                            int bcp_offset, size_t index_size) {
 744   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 745   assert_different_registers(cache, tmp);
 746   assert_not_delayed();
 747   get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size);
 748   // convert from field index to ConstantPoolCacheEntry index and from
 749   // word index to byte offset
 750   sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
 751   add(LcpoolCache, tmp, cache);
 752 }
 753 
 754 
 755 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 756                                                                         Register temp,
 757                                                                         Register bytecode,
 758                                                                         int byte_no,
 759                                                                         int bcp_offset,
 760                                                                         size_t index_size) {
 761   get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size);
 762   ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode);
 763   const int shift_count = (1 + byte_no) * BitsPerByte;
 764   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
 765          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
 766          "correct shift count");
 767   srl(bytecode, shift_count, bytecode);
 768   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
 769   and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode);
 770 }
 771 
 772 
 773 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
 774                                                                int bcp_offset, size_t index_size) {
 775   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 776   assert_different_registers(cache, tmp);
 777   assert_not_delayed();
 778   if (index_size == sizeof(u2)) {
 779     get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned);
 780   } else {
 781     ShouldNotReachHere();  // other sizes not supported here
 782   }
 783               // convert from field index to ConstantPoolCacheEntry index
 784               // and from word index to byte offset
 785   sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
 786               // skip past the header
 787   add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp);
 788               // construct pointer to cache entry
 789   add(LcpoolCache, tmp, cache);
 790 }
 791 
 792 
 793 // Load object from cpool->resolved_references(index)
 794 void InterpreterMacroAssembler::load_resolved_reference_at_index(
 795                                            Register result, Register index) {
 796   assert_different_registers(result, index);
 797   assert_not_delayed();
 798   // convert from field index to resolved_references() index and from
 799   // word index to byte offset. Since this is a java object, it can be compressed
 800   Register tmp = index;  // reuse
 801   sll(index, LogBytesPerHeapOop, tmp);
 802   get_constant_pool(result);
 803   // load pointer for resolved_references[] objArray
 804   ld_ptr(result, ConstantPool::resolved_references_offset_in_bytes(), result);
 805   // JNIHandles::resolve(result)
 806   ld_ptr(result, 0, result);
 807   // Add in the index
 808   add(result, tmp, result);
 809   load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
 810 }
 811 
 812 
 813 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
 814 // a subtype of super_klass.  Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
 815 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 816                                                   Register Rsuper_klass,
 817                                                   Register Rtmp1,
 818                                                   Register Rtmp2,
 819                                                   Register Rtmp3,
 820                                                   Label &ok_is_subtype ) {
 821   Label not_subtype;
 822 
 823   // Profile the not-null value's klass.
 824   profile_typecheck(Rsub_klass, Rtmp1);
 825 
 826   check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass,
 827                                 Rtmp1, Rtmp2,
 828                                 &ok_is_subtype, &not_subtype, NULL);
 829 
 830   check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass,
 831                                 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg,
 832                                 &ok_is_subtype, NULL);
 833 
 834   bind(not_subtype);
 835   profile_typecheck_failed(Rtmp1);
 836 }
 837 
 838 // Separate these two to allow for delay slot in middle
 839 // These are used to do a test and full jump to exception-throwing code.
 840 
 841 // %%%%% Could possibly reoptimize this by testing to see if could use
 842 // a single conditional branch (i.e. if span is small enough.
 843 // If you go that route, than get rid of the split and give up
 844 // on the delay-slot hack.
 845 
 846 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition,
 847                                                     Label&    ok ) {
 848   assert_not_delayed();
 849   br(ok_condition, true, pt, ok);
 850   // DELAY SLOT
 851 }
 852 
 853 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition,
 854                                                     Label&    ok ) {
 855   assert_not_delayed();
 856   bp( ok_condition, true, Assembler::xcc, pt, ok);
 857   // DELAY SLOT
 858 }
 859 
 860 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition,
 861                                                   Label&    ok ) {
 862   assert_not_delayed();
 863   brx(ok_condition, true, pt, ok);
 864   // DELAY SLOT
 865 }
 866 
 867 void InterpreterMacroAssembler::throw_if_not_2( address  throw_entry_point,
 868                                                 Register Rscratch,
 869                                                 Label&   ok ) {
 870   assert(throw_entry_point != NULL, "entry point must be generated by now");
 871   AddressLiteral dest(throw_entry_point);
 872   jump_to(dest, Rscratch);
 873   delayed()->nop();
 874   bind(ok);
 875 }
 876 
 877 
 878 // And if you cannot use the delay slot, here is a shorthand:
 879 
 880 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition,
 881                                                   address   throw_entry_point,
 882                                                   Register  Rscratch ) {
 883   Label ok;
 884   if (ok_condition != never) {
 885     throw_if_not_1_icc( ok_condition, ok);
 886     delayed()->nop();
 887   }
 888   throw_if_not_2( throw_entry_point, Rscratch, ok);
 889 }
 890 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition,
 891                                                   address   throw_entry_point,
 892                                                   Register  Rscratch ) {
 893   Label ok;
 894   if (ok_condition != never) {
 895     throw_if_not_1_xcc( ok_condition, ok);
 896     delayed()->nop();
 897   }
 898   throw_if_not_2( throw_entry_point, Rscratch, ok);
 899 }
 900 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition,
 901                                                 address   throw_entry_point,
 902                                                 Register  Rscratch ) {
 903   Label ok;
 904   if (ok_condition != never) {
 905     throw_if_not_1_x( ok_condition, ok);
 906     delayed()->nop();
 907   }
 908   throw_if_not_2( throw_entry_point, Rscratch, ok);
 909 }
 910 
 911 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res
 912 // Note: res is still shy of address by array offset into object.
 913 
 914 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) {
 915   assert_not_delayed();
 916 
 917   verify_oop(array);
 918 #ifdef _LP64
 919   // sign extend since tos (index) can be a 32bit value
 920   sra(index, G0, index);
 921 #endif // _LP64
 922 
 923   // check array
 924   Label ptr_ok;
 925   tst(array);
 926   throw_if_not_1_x( notZero, ptr_ok );
 927   delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index
 928   throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok);
 929 
 930   Label index_ok;
 931   cmp(index, tmp);
 932   throw_if_not_1_icc( lessUnsigned, index_ok );
 933   if (index_shift > 0)  delayed()->sll(index, index_shift, index);
 934   else                  delayed()->add(array, index, res); // addr - const offset in index
 935   // convention: move aberrant index into G3_scratch for exception message
 936   mov(index, G3_scratch);
 937   throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok);
 938 
 939   // add offset if didn't do it in delay slot
 940   if (index_shift > 0)   add(array, index, res); // addr - const offset in index
 941 }
 942 
 943 
 944 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) {
 945   assert_not_delayed();
 946 
 947   // pop array
 948   pop_ptr(array);
 949 
 950   // check array
 951   index_check_without_pop(array, index, index_shift, tmp, res);
 952 }
 953 
 954 
 955 void InterpreterMacroAssembler::get_const(Register Rdst) {
 956   ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst);
 957 }
 958 
 959 
 960 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
 961   get_const(Rdst);
 962   ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
 963 }
 964 
 965 
 966 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
 967   get_constant_pool(Rdst);
 968   ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
 969 }
 970 
 971 
 972 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
 973   get_constant_pool(Rcpool);
 974   ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags);
 975 }
 976 
 977 
 978 // unlock if synchronized method
 979 //
 980 // Unlock the receiver if this is a synchronized method.
 981 // Unlock any Java monitors from syncronized blocks.
 982 //
 983 // If there are locked Java monitors
 984 //    If throw_monitor_exception
 985 //       throws IllegalMonitorStateException
 986 //    Else if install_monitor_exception
 987 //       installs IllegalMonitorStateException
 988 //    Else
 989 //       no error processing
 990 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
 991                                                               bool throw_monitor_exception,
 992                                                               bool install_monitor_exception) {
 993   Label unlocked, unlock, no_unlock;
 994 
 995   // get the value of _do_not_unlock_if_synchronized into G1_scratch
 996   const Address do_not_unlock_if_synchronized(G2_thread,
 997     JavaThread::do_not_unlock_if_synchronized_offset());
 998   ldbool(do_not_unlock_if_synchronized, G1_scratch);
 999   stbool(G0, do_not_unlock_if_synchronized); // reset the flag
1000 
1001   // check if synchronized method
1002   const Address access_flags(Lmethod, Method::access_flags_offset());
1003   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1004   push(state); // save tos
1005   ld(access_flags, G3_scratch); // Load access flags.
1006   btst(JVM_ACC_SYNCHRONIZED, G3_scratch);
1007   br(zero, false, pt, unlocked);
1008   delayed()->nop();
1009 
1010   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1011   // is set.
1012   cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock);
1013   delayed()->nop();
1014 
1015   // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1016   // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1017 
1018   //Intel: if (throw_monitor_exception) ... else ...
1019   // Entry already unlocked, need to throw exception
1020   //...
1021 
1022   // pass top-most monitor elem
1023   add( top_most_monitor(), O1 );
1024 
1025   ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch);
1026   br_notnull_short(G3_scratch, pt, unlock);
1027 
1028   if (throw_monitor_exception) {
1029     // Entry already unlocked need to throw an exception
1030     MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1031     should_not_reach_here();
1032   } else {
1033     // Monitor already unlocked during a stack unroll.
1034     // If requested, install an illegal_monitor_state_exception.
1035     // Continue with stack unrolling.
1036     if (install_monitor_exception) {
1037       MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1038     }
1039     ba_short(unlocked);
1040   }
1041 
1042   bind(unlock);
1043 
1044   unlock_object(O1);
1045 
1046   bind(unlocked);
1047 
1048   // I0, I1: Might contain return value
1049 
1050   // Check that all monitors are unlocked
1051   { Label loop, exception, entry, restart;
1052 
1053     Register Rmptr   = O0;
1054     Register Rtemp   = O1;
1055     Register Rlimit  = Lmonitors;
1056     const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1057     assert( (delta & LongAlignmentMask) == 0,
1058             "sizeof BasicObjectLock must be even number of doublewords");
1059 
1060     #ifdef ASSERT
1061     add(top_most_monitor(), Rmptr, delta);
1062     { Label L;
1063       // ensure that Rmptr starts out above (or at) Rlimit
1064       cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1065       stop("monitor stack has negative size");
1066       bind(L);
1067     }
1068     #endif
1069     bind(restart);
1070     ba(entry);
1071     delayed()->
1072     add(top_most_monitor(), Rmptr, delta);      // points to current entry, starting with bottom-most entry
1073 
1074     // Entry is still locked, need to throw exception
1075     bind(exception);
1076     if (throw_monitor_exception) {
1077       MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1078       should_not_reach_here();
1079     } else {
1080       // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
1081       // Unlock does not block, so don't have to worry about the frame
1082       unlock_object(Rmptr);
1083       if (install_monitor_exception) {
1084         MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1085       }
1086       ba_short(restart);
1087     }
1088 
1089     bind(loop);
1090     cmp(Rtemp, G0);                             // check if current entry is used
1091     brx(Assembler::notEqual, false, pn, exception);
1092     delayed()->
1093     dec(Rmptr, delta);                          // otherwise advance to next entry
1094     #ifdef ASSERT
1095     { Label L;
1096       // ensure that Rmptr has not somehow stepped below Rlimit
1097       cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1098       stop("ran off the end of the monitor stack");
1099       bind(L);
1100     }
1101     #endif
1102     bind(entry);
1103     cmp(Rmptr, Rlimit);                         // check if bottom reached
1104     brx(Assembler::notEqual, true, pn, loop);   // if not at bottom then check this entry
1105     delayed()->
1106     ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp);
1107   }
1108 
1109   bind(no_unlock);
1110   pop(state);
1111   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1112 }
1113 
1114 
1115 // remove activation
1116 //
1117 // Unlock the receiver if this is a synchronized method.
1118 // Unlock any Java monitors from syncronized blocks.
1119 // Remove the activation from the stack.
1120 //
1121 // If there are locked Java monitors
1122 //    If throw_monitor_exception
1123 //       throws IllegalMonitorStateException
1124 //    Else if install_monitor_exception
1125 //       installs IllegalMonitorStateException
1126 //    Else
1127 //       no error processing
1128 void InterpreterMacroAssembler::remove_activation(TosState state,
1129                                                   bool throw_monitor_exception,
1130                                                   bool install_monitor_exception) {
1131 
1132   unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
1133 
1134   // save result (push state before jvmti call and pop it afterwards) and notify jvmti
1135   notify_method_exit(false, state, NotifyJVMTI);
1136 
1137   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1138   verify_thread();
1139 
1140   // return tos
1141   assert(Otos_l1 == Otos_i, "adjust code below");
1142   switch (state) {
1143 #ifdef _LP64
1144   case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0
1145 #else
1146   case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through  // O1 -> I1
1147 #endif
1148   case btos:                                      // fall through
1149   case ctos:
1150   case stos:                                      // fall through
1151   case atos:                                      // fall through
1152   case itos: mov(Otos_l1, Otos_l1->after_save());    break;        // O0 -> I0
1153   case ftos:                                      // fall through
1154   case dtos:                                      // fall through
1155   case vtos: /* nothing to do */                     break;
1156   default  : ShouldNotReachHere();
1157   }
1158 
1159 #if defined(COMPILER2) && !defined(_LP64)
1160   if (state == ltos) {
1161     // C2 expects long results in G1 we can't tell if we're returning to interpreted
1162     // or compiled so just be safe use G1 and O0/O1
1163 
1164     // Shift bits into high (msb) of G1
1165     sllx(Otos_l1->after_save(), 32, G1);
1166     // Zero extend low bits
1167     srl (Otos_l2->after_save(), 0, Otos_l2->after_save());
1168     or3 (Otos_l2->after_save(), G1, G1);
1169   }
1170 #endif /* COMPILER2 */
1171 
1172 }
1173 #endif /* CC_INTERP */
1174 
1175 
1176 // Lock object
1177 //
1178 // Argument - lock_reg points to the BasicObjectLock to be used for locking,
1179 //            it must be initialized with the object to lock
1180 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) {
1181   if (UseHeavyMonitors) {
1182     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1183   }
1184   else {
1185     Register obj_reg = Object;
1186     Register mark_reg = G4_scratch;
1187     Register temp_reg = G1_scratch;
1188     Address  lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes());
1189     Address  mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1190     Label    done;
1191 
1192     Label slow_case;
1193 
1194     assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg);
1195 
1196     // load markOop from object into mark_reg
1197     ld_ptr(mark_addr, mark_reg);
1198 
1199     if (UseBiasedLocking) {
1200       biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case);
1201     }
1202 
1203     // get the address of basicLock on stack that will be stored in the object
1204     // we need a temporary register here as we do not want to clobber lock_reg
1205     // (cas clobbers the destination register)
1206     mov(lock_reg, temp_reg);
1207     // set mark reg to be (markOop of object | UNLOCK_VALUE)
1208     or3(mark_reg, markOopDesc::unlocked_value, mark_reg);
1209     // initialize the box  (Must happen before we update the object mark!)
1210     st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1211     // compare and exchange object_addr, markOop | 1, stack address of basicLock
1212     assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1213     casx_under_lock(mark_addr.base(), mark_reg, temp_reg,
1214       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
1215 
1216     // if the compare and exchange succeeded we are done (we saw an unlocked object)
1217     cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done);
1218 
1219     // We did not see an unlocked object so try the fast recursive case
1220 
1221     // Check if owner is self by comparing the value in the markOop of object
1222     // with the stack pointer
1223     sub(temp_reg, SP, temp_reg);
1224 #ifdef _LP64
1225     sub(temp_reg, STACK_BIAS, temp_reg);
1226 #endif
1227     assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1228 
1229     // Composite "andcc" test:
1230     // (a) %sp -vs- markword proximity check, and,
1231     // (b) verify mark word LSBs == 0 (Stack-locked).
1232     //
1233     // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size())
1234     // Note that the page size used for %sp proximity testing is arbitrary and is
1235     // unrelated to the actual MMU page size.  We use a 'logical' page size of
1236     // 4096 bytes.   F..FFF003 is designed to fit conveniently in the SIMM13 immediate
1237     // field of the andcc instruction.
1238     andcc (temp_reg, 0xFFFFF003, G0) ;
1239 
1240     // if condition is true we are done and hence we can store 0 in the displaced
1241     // header indicating it is a recursive lock and be done
1242     brx(Assembler::zero, true, Assembler::pt, done);
1243     delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1244 
1245     // none of the above fast optimizations worked so we have to get into the
1246     // slow case of monitor enter
1247     bind(slow_case);
1248     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1249 
1250     bind(done);
1251   }
1252 }
1253 
1254 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1255 //
1256 // Argument - lock_reg points to the BasicObjectLock for lock
1257 // Throw IllegalMonitorException if object is not locked by current thread
1258 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1259   if (UseHeavyMonitors) {
1260     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1261   } else {
1262     Register obj_reg = G3_scratch;
1263     Register mark_reg = G4_scratch;
1264     Register displaced_header_reg = G1_scratch;
1265     Address  lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes());
1266     Address  mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1267     Label    done;
1268 
1269     if (UseBiasedLocking) {
1270       // load the object out of the BasicObjectLock
1271       ld_ptr(lockobj_addr, obj_reg);
1272       biased_locking_exit(mark_addr, mark_reg, done, true);
1273       st_ptr(G0, lockobj_addr);  // free entry
1274     }
1275 
1276     // Test first if we are in the fast recursive case
1277     Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes());
1278     ld_ptr(lock_addr, displaced_header_reg);
1279     br_null(displaced_header_reg, true, Assembler::pn, done);
1280     delayed()->st_ptr(G0, lockobj_addr);  // free entry
1281 
1282     // See if it is still a light weight lock, if so we just unlock
1283     // the object and we are done
1284 
1285     if (!UseBiasedLocking) {
1286       // load the object out of the BasicObjectLock
1287       ld_ptr(lockobj_addr, obj_reg);
1288     }
1289 
1290     // we have the displaced header in displaced_header_reg
1291     // we expect to see the stack address of the basicLock in case the
1292     // lock is still a light weight lock (lock_reg)
1293     assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1294     casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg,
1295       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
1296     cmp(lock_reg, displaced_header_reg);
1297     brx(Assembler::equal, true, Assembler::pn, done);
1298     delayed()->st_ptr(G0, lockobj_addr);  // free entry
1299 
1300     // The lock has been converted into a heavy lock and hence
1301     // we need to get into the slow case
1302 
1303     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1304 
1305     bind(done);
1306   }
1307 }
1308 
1309 #ifndef CC_INTERP
1310 
1311 // Get the method data pointer from the Method* and set the
1312 // specified register to its value.
1313 
1314 void InterpreterMacroAssembler::set_method_data_pointer() {
1315   assert(ProfileInterpreter, "must be profiling interpreter");
1316   Label get_continue;
1317 
1318   ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1319   test_method_data_pointer(get_continue);
1320   add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1321   bind(get_continue);
1322 }
1323 
1324 // Set the method data pointer for the current bcp.
1325 
1326 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1327   assert(ProfileInterpreter, "must be profiling interpreter");
1328   Label zero_continue;
1329 
1330   // Test MDO to avoid the call if it is NULL.
1331   ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1332   test_method_data_pointer(zero_continue);
1333   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
1334   add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1335   add(ImethodDataPtr, O0, ImethodDataPtr);
1336   bind(zero_continue);
1337 }
1338 
1339 // Test ImethodDataPtr.  If it is null, continue at the specified label
1340 
1341 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1342   assert(ProfileInterpreter, "must be profiling interpreter");
1343   br_null_short(ImethodDataPtr, Assembler::pn, zero_continue);
1344 }
1345 
1346 void InterpreterMacroAssembler::verify_method_data_pointer() {
1347   assert(ProfileInterpreter, "must be profiling interpreter");
1348 #ifdef ASSERT
1349   Label verify_continue;
1350   test_method_data_pointer(verify_continue);
1351 
1352   // If the mdp is valid, it will point to a DataLayout header which is
1353   // consistent with the bcp.  The converse is highly probable also.
1354   lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch);
1355   ld_ptr(Lmethod, Method::const_offset(), O5);
1356   add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch);
1357   add(G3_scratch, O5, G3_scratch);
1358   cmp(Lbcp, G3_scratch);
1359   brx(Assembler::equal, false, Assembler::pt, verify_continue);
1360 
1361   Register temp_reg = O5;
1362   delayed()->mov(ImethodDataPtr, temp_reg);
1363   // %%% should use call_VM_leaf here?
1364   //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr);
1365   save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1);
1366   Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
1367   stf(FloatRegisterImpl::D, Ftos_d, d_save);
1368   mov(temp_reg->after_save(), O2);
1369   save_thread(L7_thread_cache);
1370   call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none);
1371   delayed()->nop();
1372   restore_thread(L7_thread_cache);
1373   ldf(FloatRegisterImpl::D, d_save, Ftos_d);
1374   restore();
1375   bind(verify_continue);
1376 #endif // ASSERT
1377 }
1378 
1379 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1380                                                                 Register Rtmp,
1381                                                                 Label &profile_continue) {
1382   assert(ProfileInterpreter, "must be profiling interpreter");
1383   // Control will flow to "profile_continue" if the counter is less than the
1384   // limit or if we call profile_method()
1385 
1386   Label done;
1387 
1388   // if no method data exists, and the counter is high enough, make one
1389   br_notnull_short(ImethodDataPtr, Assembler::pn, done);
1390 
1391   // Test to see if we should create a method data oop
1392   AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit);
1393   sethi(profile_limit, Rtmp);
1394   ld(Rtmp, profile_limit.low10(), Rtmp);
1395   cmp(invocation_count, Rtmp);
1396   // Use long branches because call_VM() code and following code generated by
1397   // test_backedge_count_for_osr() is large in debug VM.
1398   br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue);
1399   delayed()->nop();
1400 
1401   // Build it now.
1402   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1403   set_method_data_pointer_for_bcp();
1404   ba(profile_continue);
1405   delayed()->nop();
1406   bind(done);
1407 }
1408 
1409 // Store a value at some constant offset from the method data pointer.
1410 
1411 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1412   assert(ProfileInterpreter, "must be profiling interpreter");
1413   st_ptr(value, ImethodDataPtr, constant);
1414 }
1415 
1416 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter,
1417                                                       Register bumped_count,
1418                                                       bool decrement) {
1419   assert(ProfileInterpreter, "must be profiling interpreter");
1420 
1421   // Load the counter.
1422   ld_ptr(counter, bumped_count);
1423 
1424   if (decrement) {
1425     // Decrement the register.  Set condition codes.
1426     subcc(bumped_count, DataLayout::counter_increment, bumped_count);
1427 
1428     // If the decrement causes the counter to overflow, stay negative
1429     Label L;
1430     brx(Assembler::negative, true, Assembler::pn, L);
1431 
1432     // Store the decremented counter, if it is still negative.
1433     delayed()->st_ptr(bumped_count, counter);
1434     bind(L);
1435   } else {
1436     // Increment the register.  Set carry flag.
1437     addcc(bumped_count, DataLayout::counter_increment, bumped_count);
1438 
1439     // If the increment causes the counter to overflow, pull back by 1.
1440     assert(DataLayout::counter_increment == 1, "subc works");
1441     subc(bumped_count, G0, bumped_count);
1442 
1443     // Store the incremented counter.
1444     st_ptr(bumped_count, counter);
1445   }
1446 }
1447 
1448 // Increment the value at some constant offset from the method data pointer.
1449 
1450 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1451                                                       Register bumped_count,
1452                                                       bool decrement) {
1453   // Locate the counter at a fixed offset from the mdp:
1454   Address counter(ImethodDataPtr, constant);
1455   increment_mdp_data_at(counter, bumped_count, decrement);
1456 }
1457 
1458 // Increment the value at some non-fixed (reg + constant) offset from
1459 // the method data pointer.
1460 
1461 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1462                                                       int constant,
1463                                                       Register bumped_count,
1464                                                       Register scratch2,
1465                                                       bool decrement) {
1466   // Add the constant to reg to get the offset.
1467   add(ImethodDataPtr, reg, scratch2);
1468   Address counter(scratch2, constant);
1469   increment_mdp_data_at(counter, bumped_count, decrement);
1470 }
1471 
1472 // Set a flag value at the current method data pointer position.
1473 // Updates a single byte of the header, to avoid races with other header bits.
1474 
1475 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1476                                                 Register scratch) {
1477   assert(ProfileInterpreter, "must be profiling interpreter");
1478   // Load the data header
1479   ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch);
1480 
1481   // Set the flag
1482   or3(scratch, flag_constant, scratch);
1483 
1484   // Store the modified header.
1485   stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset()));
1486 }
1487 
1488 // Test the location at some offset from the method data pointer.
1489 // If it is not equal to value, branch to the not_equal_continue Label.
1490 // Set condition codes to match the nullness of the loaded value.
1491 
1492 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1493                                                  Register value,
1494                                                  Label& not_equal_continue,
1495                                                  Register scratch) {
1496   assert(ProfileInterpreter, "must be profiling interpreter");
1497   ld_ptr(ImethodDataPtr, offset, scratch);
1498   cmp(value, scratch);
1499   brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue);
1500   delayed()->tst(scratch);
1501 }
1502 
1503 // Update the method data pointer by the displacement located at some fixed
1504 // offset from the method data pointer.
1505 
1506 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1507                                                      Register scratch) {
1508   assert(ProfileInterpreter, "must be profiling interpreter");
1509   ld_ptr(ImethodDataPtr, offset_of_disp, scratch);
1510   add(ImethodDataPtr, scratch, ImethodDataPtr);
1511 }
1512 
1513 // Update the method data pointer by the displacement located at the
1514 // offset (reg + offset_of_disp).
1515 
1516 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1517                                                      int offset_of_disp,
1518                                                      Register scratch) {
1519   assert(ProfileInterpreter, "must be profiling interpreter");
1520   add(reg, offset_of_disp, scratch);
1521   ld_ptr(ImethodDataPtr, scratch, scratch);
1522   add(ImethodDataPtr, scratch, ImethodDataPtr);
1523 }
1524 
1525 // Update the method data pointer by a simple constant displacement.
1526 
1527 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1528   assert(ProfileInterpreter, "must be profiling interpreter");
1529   add(ImethodDataPtr, constant, ImethodDataPtr);
1530 }
1531 
1532 // Update the method data pointer for a _ret bytecode whose target
1533 // was not among our cached targets.
1534 
1535 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1536                                                    Register return_bci) {
1537   assert(ProfileInterpreter, "must be profiling interpreter");
1538   push(state);
1539   st_ptr(return_bci, l_tmp);  // protect return_bci, in case it is volatile
1540   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1541   ld_ptr(l_tmp, return_bci);
1542   pop(state);
1543 }
1544 
1545 // Count a taken branch in the bytecodes.
1546 
1547 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1548   if (ProfileInterpreter) {
1549     Label profile_continue;
1550 
1551     // If no method data exists, go to profile_continue.
1552     test_method_data_pointer(profile_continue);
1553 
1554     // We are taking a branch.  Increment the taken count.
1555     increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count);
1556 
1557     // The method data pointer needs to be updated to reflect the new target.
1558     update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1559     bind (profile_continue);
1560   }
1561 }
1562 
1563 
1564 // Count a not-taken branch in the bytecodes.
1565 
1566 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) {
1567   if (ProfileInterpreter) {
1568     Label profile_continue;
1569 
1570     // If no method data exists, go to profile_continue.
1571     test_method_data_pointer(profile_continue);
1572 
1573     // We are taking a branch.  Increment the not taken count.
1574     increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch);
1575 
1576     // The method data pointer needs to be updated to correspond to the
1577     // next bytecode.
1578     update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1579     bind (profile_continue);
1580   }
1581 }
1582 
1583 
1584 // Count a non-virtual call in the bytecodes.
1585 
1586 void InterpreterMacroAssembler::profile_call(Register scratch) {
1587   if (ProfileInterpreter) {
1588     Label profile_continue;
1589 
1590     // If no method data exists, go to profile_continue.
1591     test_method_data_pointer(profile_continue);
1592 
1593     // We are making a call.  Increment the count.
1594     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1595 
1596     // The method data pointer needs to be updated to reflect the new target.
1597     update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1598     bind (profile_continue);
1599   }
1600 }
1601 
1602 
1603 // Count a final call in the bytecodes.
1604 
1605 void InterpreterMacroAssembler::profile_final_call(Register scratch) {
1606   if (ProfileInterpreter) {
1607     Label profile_continue;
1608 
1609     // If no method data exists, go to profile_continue.
1610     test_method_data_pointer(profile_continue);
1611 
1612     // We are making a call.  Increment the count.
1613     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1614 
1615     // The method data pointer needs to be updated to reflect the new target.
1616     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1617     bind (profile_continue);
1618   }
1619 }
1620 
1621 
1622 // Count a virtual call in the bytecodes.
1623 
1624 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1625                                                      Register scratch,
1626                                                      bool receiver_can_be_null) {
1627   if (ProfileInterpreter) {
1628     Label profile_continue;
1629 
1630     // If no method data exists, go to profile_continue.
1631     test_method_data_pointer(profile_continue);
1632 
1633 
1634     Label skip_receiver_profile;
1635     if (receiver_can_be_null) {
1636       Label not_null;
1637       br_notnull_short(receiver, Assembler::pt, not_null);
1638       // We are making a call.  Increment the count for null receiver.
1639       increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1640       ba_short(skip_receiver_profile);
1641       bind(not_null);
1642     }
1643 
1644     // Record the receiver type.
1645     record_klass_in_profile(receiver, scratch, true);
1646     bind(skip_receiver_profile);
1647 
1648     // The method data pointer needs to be updated to reflect the new target.
1649     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1650     bind (profile_continue);
1651   }
1652 }
1653 
1654 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1655                                         Register receiver, Register scratch,
1656                                         int start_row, Label& done, bool is_virtual_call) {
1657   if (TypeProfileWidth == 0) {
1658     if (is_virtual_call) {
1659       increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1660     }
1661     return;
1662   }
1663 
1664   int last_row = VirtualCallData::row_limit() - 1;
1665   assert(start_row <= last_row, "must be work left to do");
1666   // Test this row for both the receiver and for null.
1667   // Take any of three different outcomes:
1668   //   1. found receiver => increment count and goto done
1669   //   2. found null => keep looking for case 1, maybe allocate this cell
1670   //   3. found something else => keep looking for cases 1 and 2
1671   // Case 3 is handled by a recursive call.
1672   for (int row = start_row; row <= last_row; row++) {
1673     Label next_test;
1674     bool test_for_null_also = (row == start_row);
1675 
1676     // See if the receiver is receiver[n].
1677     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1678     test_mdp_data_at(recvr_offset, receiver, next_test, scratch);
1679     // delayed()->tst(scratch);
1680 
1681     // The receiver is receiver[n].  Increment count[n].
1682     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1683     increment_mdp_data_at(count_offset, scratch);
1684     ba_short(done);
1685     bind(next_test);
1686 
1687     if (test_for_null_also) {
1688       Label found_null;
1689       // Failed the equality check on receiver[n]...  Test for null.
1690       if (start_row == last_row) {
1691         // The only thing left to do is handle the null case.
1692         if (is_virtual_call) {
1693           brx(Assembler::zero, false, Assembler::pn, found_null);
1694           delayed()->nop();
1695           // Receiver did not match any saved receiver and there is no empty row for it.
1696           // Increment total counter to indicate polymorphic case.
1697           increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1698           ba_short(done);
1699           bind(found_null);
1700         } else {
1701           brx(Assembler::notZero, false, Assembler::pt, done);
1702           delayed()->nop();
1703         }
1704         break;
1705       }
1706       // Since null is rare, make it be the branch-taken case.
1707       brx(Assembler::zero, false, Assembler::pn, found_null);
1708       delayed()->nop();
1709 
1710       // Put all the "Case 3" tests here.
1711       record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call);
1712 
1713       // Found a null.  Keep searching for a matching receiver,
1714       // but remember that this is an empty (unused) slot.
1715       bind(found_null);
1716     }
1717   }
1718 
1719   // In the fall-through case, we found no matching receiver, but we
1720   // observed the receiver[start_row] is NULL.
1721 
1722   // Fill in the receiver field and increment the count.
1723   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1724   set_mdp_data_at(recvr_offset, receiver);
1725   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1726   mov(DataLayout::counter_increment, scratch);
1727   set_mdp_data_at(count_offset, scratch);
1728   if (start_row > 0) {
1729     ba_short(done);
1730   }
1731 }
1732 
1733 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1734                                                         Register scratch, bool is_virtual_call) {
1735   assert(ProfileInterpreter, "must be profiling");
1736   Label done;
1737 
1738   record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call);
1739 
1740   bind (done);
1741 }
1742 
1743 
1744 // Count a ret in the bytecodes.
1745 
1746 void InterpreterMacroAssembler::profile_ret(TosState state,
1747                                             Register return_bci,
1748                                             Register scratch) {
1749   if (ProfileInterpreter) {
1750     Label profile_continue;
1751     uint row;
1752 
1753     // If no method data exists, go to profile_continue.
1754     test_method_data_pointer(profile_continue);
1755 
1756     // Update the total ret count.
1757     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1758 
1759     for (row = 0; row < RetData::row_limit(); row++) {
1760       Label next_test;
1761 
1762       // See if return_bci is equal to bci[n]:
1763       test_mdp_data_at(in_bytes(RetData::bci_offset(row)),
1764                        return_bci, next_test, scratch);
1765 
1766       // return_bci is equal to bci[n].  Increment the count.
1767       increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch);
1768 
1769       // The method data pointer needs to be updated to reflect the new target.
1770       update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch);
1771       ba_short(profile_continue);
1772       bind(next_test);
1773     }
1774 
1775     update_mdp_for_ret(state, return_bci);
1776 
1777     bind (profile_continue);
1778   }
1779 }
1780 
1781 // Profile an unexpected null in the bytecodes.
1782 void InterpreterMacroAssembler::profile_null_seen(Register scratch) {
1783   if (ProfileInterpreter) {
1784     Label profile_continue;
1785 
1786     // If no method data exists, go to profile_continue.
1787     test_method_data_pointer(profile_continue);
1788 
1789     set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch);
1790 
1791     // The method data pointer needs to be updated.
1792     int mdp_delta = in_bytes(BitData::bit_data_size());
1793     if (TypeProfileCasts) {
1794       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1795     }
1796     update_mdp_by_constant(mdp_delta);
1797 
1798     bind (profile_continue);
1799   }
1800 }
1801 
1802 void InterpreterMacroAssembler::profile_typecheck(Register klass,
1803                                                   Register scratch) {
1804   if (ProfileInterpreter) {
1805     Label profile_continue;
1806 
1807     // If no method data exists, go to profile_continue.
1808     test_method_data_pointer(profile_continue);
1809 
1810     int mdp_delta = in_bytes(BitData::bit_data_size());
1811     if (TypeProfileCasts) {
1812       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1813 
1814       // Record the object type.
1815       record_klass_in_profile(klass, scratch, false);
1816     }
1817 
1818     // The method data pointer needs to be updated.
1819     update_mdp_by_constant(mdp_delta);
1820 
1821     bind (profile_continue);
1822   }
1823 }
1824 
1825 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) {
1826   if (ProfileInterpreter && TypeProfileCasts) {
1827     Label profile_continue;
1828 
1829     // If no method data exists, go to profile_continue.
1830     test_method_data_pointer(profile_continue);
1831 
1832     int count_offset = in_bytes(CounterData::count_offset());
1833     // Back up the address, since we have already bumped the mdp.
1834     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1835 
1836     // *Decrement* the counter.  We expect to see zero or small negatives.
1837     increment_mdp_data_at(count_offset, scratch, true);
1838 
1839     bind (profile_continue);
1840   }
1841 }
1842 
1843 // Count the default case of a switch construct.
1844 
1845 void InterpreterMacroAssembler::profile_switch_default(Register scratch) {
1846   if (ProfileInterpreter) {
1847     Label profile_continue;
1848 
1849     // If no method data exists, go to profile_continue.
1850     test_method_data_pointer(profile_continue);
1851 
1852     // Update the default case count
1853     increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1854                           scratch);
1855 
1856     // The method data pointer needs to be updated.
1857     update_mdp_by_offset(
1858                     in_bytes(MultiBranchData::default_displacement_offset()),
1859                     scratch);
1860 
1861     bind (profile_continue);
1862   }
1863 }
1864 
1865 // Count the index'th case of a switch construct.
1866 
1867 void InterpreterMacroAssembler::profile_switch_case(Register index,
1868                                                     Register scratch,
1869                                                     Register scratch2,
1870                                                     Register scratch3) {
1871   if (ProfileInterpreter) {
1872     Label profile_continue;
1873 
1874     // If no method data exists, go to profile_continue.
1875     test_method_data_pointer(profile_continue);
1876 
1877     // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1878     set(in_bytes(MultiBranchData::per_case_size()), scratch);
1879     smul(index, scratch, scratch);
1880     add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch);
1881 
1882     // Update the case count
1883     increment_mdp_data_at(scratch,
1884                           in_bytes(MultiBranchData::relative_count_offset()),
1885                           scratch2,
1886                           scratch3);
1887 
1888     // The method data pointer needs to be updated.
1889     update_mdp_by_offset(scratch,
1890                      in_bytes(MultiBranchData::relative_displacement_offset()),
1891                      scratch2);
1892 
1893     bind (profile_continue);
1894   }
1895 }
1896 
1897 // add a InterpMonitorElem to stack (see frame_sparc.hpp)
1898 
1899 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty,
1900                                                       Register Rtemp,
1901                                                       Register Rtemp2 ) {
1902 
1903   Register Rlimit = Lmonitors;
1904   const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1905   assert( (delta & LongAlignmentMask) == 0,
1906           "sizeof BasicObjectLock must be even number of doublewords");
1907 
1908   sub( SP,        delta, SP);
1909   sub( Lesp,      delta, Lesp);
1910   sub( Lmonitors, delta, Lmonitors);
1911 
1912   if (!stack_is_empty) {
1913 
1914     // must copy stack contents down
1915 
1916     Label start_copying, next;
1917 
1918     // untested("monitor stack expansion");
1919     compute_stack_base(Rtemp);
1920     ba(start_copying);
1921     delayed()->cmp(Rtemp, Rlimit); // done? duplicated below
1922 
1923     // note: must copy from low memory upwards
1924     // On entry to loop,
1925     // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS)
1926     // Loop mutates Rtemp
1927 
1928     bind( next);
1929 
1930     st_ptr(Rtemp2, Rtemp, 0);
1931     inc(Rtemp, wordSize);
1932     cmp(Rtemp, Rlimit); // are we done? (duplicated above)
1933 
1934     bind( start_copying );
1935 
1936     brx( notEqual, true, pn, next );
1937     delayed()->ld_ptr( Rtemp, delta, Rtemp2 );
1938 
1939     // done copying stack
1940   }
1941 }
1942 
1943 // Locals
1944 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) {
1945   assert_not_delayed();
1946   sll(index, Interpreter::logStackElementSize, index);
1947   sub(Llocals, index, index);
1948   ld_ptr(index, 0, dst);
1949   // Note:  index must hold the effective address--the iinc template uses it
1950 }
1951 
1952 // Just like access_local_ptr but the tag is a returnAddress
1953 void InterpreterMacroAssembler::access_local_returnAddress(Register index,
1954                                                            Register dst ) {
1955   assert_not_delayed();
1956   sll(index, Interpreter::logStackElementSize, index);
1957   sub(Llocals, index, index);
1958   ld_ptr(index, 0, dst);
1959 }
1960 
1961 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) {
1962   assert_not_delayed();
1963   sll(index, Interpreter::logStackElementSize, index);
1964   sub(Llocals, index, index);
1965   ld(index, 0, dst);
1966   // Note:  index must hold the effective address--the iinc template uses it
1967 }
1968 
1969 
1970 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) {
1971   assert_not_delayed();
1972   sll(index, Interpreter::logStackElementSize, index);
1973   sub(Llocals, index, index);
1974   // First half stored at index n+1 (which grows down from Llocals[n])
1975   load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst);
1976 }
1977 
1978 
1979 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) {
1980   assert_not_delayed();
1981   sll(index, Interpreter::logStackElementSize, index);
1982   sub(Llocals, index, index);
1983   ldf(FloatRegisterImpl::S, index, 0, dst);
1984 }
1985 
1986 
1987 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) {
1988   assert_not_delayed();
1989   sll(index, Interpreter::logStackElementSize, index);
1990   sub(Llocals, index, index);
1991   load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst);
1992 }
1993 
1994 
1995 #ifdef ASSERT
1996 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) {
1997   Label L;
1998 
1999   assert(Rindex != Rscratch, "Registers cannot be same");
2000   assert(Rindex != Rscratch1, "Registers cannot be same");
2001   assert(Rlimit != Rscratch, "Registers cannot be same");
2002   assert(Rlimit != Rscratch1, "Registers cannot be same");
2003   assert(Rscratch1 != Rscratch, "Registers cannot be same");
2004 
2005   // untested("reg area corruption");
2006   add(Rindex, offset, Rscratch);
2007   add(Rlimit, 64 + STACK_BIAS, Rscratch1);
2008   cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L);
2009   stop("regsave area is being clobbered");
2010   bind(L);
2011 }
2012 #endif // ASSERT
2013 
2014 
2015 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) {
2016   assert_not_delayed();
2017   sll(index, Interpreter::logStackElementSize, index);
2018   sub(Llocals, index, index);
2019   debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);)
2020   st(src, index, 0);
2021 }
2022 
2023 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) {
2024   assert_not_delayed();
2025   sll(index, Interpreter::logStackElementSize, index);
2026   sub(Llocals, index, index);
2027 #ifdef ASSERT
2028   check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2029 #endif
2030   st_ptr(src, index, 0);
2031 }
2032 
2033 
2034 
2035 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) {
2036   st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n));
2037 }
2038 
2039 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) {
2040   assert_not_delayed();
2041   sll(index, Interpreter::logStackElementSize, index);
2042   sub(Llocals, index, index);
2043 #ifdef ASSERT
2044   check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2045 #endif
2046   store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1
2047 }
2048 
2049 
2050 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) {
2051   assert_not_delayed();
2052   sll(index, Interpreter::logStackElementSize, index);
2053   sub(Llocals, index, index);
2054 #ifdef ASSERT
2055   check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2056 #endif
2057   stf(FloatRegisterImpl::S, src, index, 0);
2058 }
2059 
2060 
2061 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) {
2062   assert_not_delayed();
2063   sll(index, Interpreter::logStackElementSize, index);
2064   sub(Llocals, index, index);
2065 #ifdef ASSERT
2066   check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2067 #endif
2068   store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1));
2069 }
2070 
2071 
2072 int InterpreterMacroAssembler::top_most_monitor_byte_offset() {
2073   const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2074   int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong);
2075   return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS;
2076 }
2077 
2078 
2079 Address InterpreterMacroAssembler::top_most_monitor() {
2080   return Address(FP, top_most_monitor_byte_offset());
2081 }
2082 
2083 
2084 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) {
2085   add( Lesp,      wordSize,                                    Rdest );
2086 }
2087 
2088 #endif /* CC_INTERP */
2089 
2090 void InterpreterMacroAssembler::get_method_counters(Register method,
2091                                                     Register Rcounters,
2092                                                     Label& skip) {
2093   Label has_counters;
2094   Address method_counters(method, in_bytes(Method::method_counters_offset()));
2095   ld_ptr(method_counters, Rcounters);
2096   br_notnull_short(Rcounters, Assembler::pt, has_counters);
2097   call_VM(noreg, CAST_FROM_FN_PTR(address,
2098           InterpreterRuntime::build_method_counters), method);
2099   ld_ptr(method_counters, Rcounters);
2100   br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory
2101   delayed()->nop();
2102   bind(has_counters);
2103 }
2104 
2105 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2106   assert(UseCompiler, "incrementing must be useful");
2107   assert_different_registers(Rcounters, Rtmp, Rtmp2);
2108 
2109   Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2110                                  InvocationCounter::counter_offset());
2111   Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2112                                  InvocationCounter::counter_offset());
2113   int delta = InvocationCounter::count_increment;
2114 
2115   // Load each counter in a register
2116   ld( inv_counter, Rtmp );
2117   ld( be_counter, Rtmp2 );
2118 
2119   assert( is_simm13( delta ), " delta too large.");
2120 
2121   // Add the delta to the invocation counter and store the result
2122   add( Rtmp, delta, Rtmp );
2123 
2124   // Mask the backedge counter
2125   and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2126 
2127   // Store value
2128   st( Rtmp, inv_counter);
2129 
2130   // Add invocation counter + backedge counter
2131   add( Rtmp, Rtmp2, Rtmp);
2132 
2133   // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2134 }
2135 
2136 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2137   assert(UseCompiler, "incrementing must be useful");
2138   assert_different_registers(Rcounters, Rtmp, Rtmp2);
2139 
2140   Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2141                                  InvocationCounter::counter_offset());
2142   Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2143                                  InvocationCounter::counter_offset());
2144 
2145   int delta = InvocationCounter::count_increment;
2146   // Load each counter in a register
2147   ld( be_counter, Rtmp );
2148   ld( inv_counter, Rtmp2 );
2149 
2150   // Add the delta to the backedge counter
2151   add( Rtmp, delta, Rtmp );
2152 
2153   // Mask the invocation counter, add to backedge counter
2154   and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2155 
2156   // and store the result to memory
2157   st( Rtmp, be_counter );
2158 
2159   // Add backedge + invocation counter
2160   add( Rtmp, Rtmp2, Rtmp );
2161 
2162   // Note that this macro must leave backedge_count + invocation_count in Rtmp!
2163 }
2164 
2165 #ifndef CC_INTERP
2166 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count,
2167                                                              Register branch_bcp,
2168                                                              Register Rtmp ) {
2169   Label did_not_overflow;
2170   Label overflow_with_error;
2171   assert_different_registers(backedge_count, Rtmp, branch_bcp);
2172   assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
2173 
2174   AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit);
2175   load_contents(limit, Rtmp);
2176   cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow);
2177 
2178   // When ProfileInterpreter is on, the backedge_count comes from the
2179   // MethodData*, which value does not get reset on the call to
2180   // frequency_counter_overflow().  To avoid excessive calls to the overflow
2181   // routine while the method is being compiled, add a second test to make sure
2182   // the overflow function is called only once every overflow_frequency.
2183   if (ProfileInterpreter) {
2184     const int overflow_frequency = 1024;
2185     andcc(backedge_count, overflow_frequency-1, Rtmp);
2186     brx(Assembler::notZero, false, Assembler::pt, did_not_overflow);
2187     delayed()->nop();
2188   }
2189 
2190   // overflow in loop, pass branch bytecode
2191   set(6,Rtmp);
2192   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp);
2193 
2194   // Was an OSR adapter generated?
2195   // O0 = osr nmethod
2196   br_null_short(O0, Assembler::pn, overflow_with_error);
2197 
2198   // Has the nmethod been invalidated already?
2199   ld(O0, nmethod::entry_bci_offset(), O2);
2200   cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, overflow_with_error);
2201 
2202   // migrate the interpreter frame off of the stack
2203 
2204   mov(G2_thread, L7);
2205   // save nmethod
2206   mov(O0, L6);
2207   set_last_Java_frame(SP, noreg);
2208   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
2209   reset_last_Java_frame();
2210   mov(L7, G2_thread);
2211 
2212   // move OSR nmethod to I1
2213   mov(L6, I1);
2214 
2215   // OSR buffer to I0
2216   mov(O0, I0);
2217 
2218   // remove the interpreter frame
2219   restore(I5_savedSP, 0, SP);
2220 
2221   // Jump to the osr code.
2222   ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
2223   jmp(O2, G0);
2224   delayed()->nop();
2225 
2226   bind(overflow_with_error);
2227 
2228   bind(did_not_overflow);
2229 }
2230 
2231 
2232 
2233 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) {
2234   if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); }
2235 }
2236 
2237 
2238 // local helper function for the verify_oop_or_return_address macro
2239 static bool verify_return_address(Method* m, int bci) {
2240 #ifndef PRODUCT
2241   address pc = (address)(m->constMethod())
2242              + in_bytes(ConstMethod::codes_offset()) + bci;
2243   // assume it is a valid return address if it is inside m and is preceded by a jsr
2244   if (!m->contains(pc))                                          return false;
2245   address jsr_pc;
2246   jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2247   if (*jsr_pc == Bytecodes::_jsr   && jsr_pc >= m->code_base())    return true;
2248   jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2249   if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base())    return true;
2250 #endif // PRODUCT
2251   return false;
2252 }
2253 
2254 
2255 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2256   if (!VerifyOops)  return;
2257   // the VM documentation for the astore[_wide] bytecode allows
2258   // the TOS to be not only an oop but also a return address
2259   Label test;
2260   Label skip;
2261   // See if it is an address (in the current method):
2262 
2263   mov(reg, Rtmp);
2264   const int log2_bytecode_size_limit = 16;
2265   srl(Rtmp, log2_bytecode_size_limit, Rtmp);
2266   br_notnull_short( Rtmp, pt, test );
2267 
2268   // %%% should use call_VM_leaf here?
2269   save_frame_and_mov(0, Lmethod, O0, reg, O1);
2270   save_thread(L7_thread_cache);
2271   call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none);
2272   delayed()->nop();
2273   restore_thread(L7_thread_cache);
2274   br_notnull( O0, false, pt, skip );
2275   delayed()->restore();
2276 
2277   // Perform a more elaborate out-of-line call
2278   // Not an address; verify it:
2279   bind(test);
2280   verify_oop(reg);
2281   bind(skip);
2282 }
2283 
2284 
2285 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2286   if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
2287 }
2288 #endif /* CC_INTERP */
2289 
2290 // Inline assembly for:
2291 //
2292 // if (thread is in interp_only_mode) {
2293 //   InterpreterRuntime::post_method_entry();
2294 // }
2295 // if (DTraceMethodProbes) {
2296 //   SharedRuntime::dtrace_method_entry(method, receiver);
2297 // }
2298 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2299 //   SharedRuntime::rc_trace_method_entry(method, receiver);
2300 // }
2301 
2302 void InterpreterMacroAssembler::notify_method_entry() {
2303 
2304   // C++ interpreter only uses this for native methods.
2305 
2306   // Whenever JVMTI puts a thread in interp_only_mode, method
2307   // entry/exit events are sent for that thread to track stack
2308   // depth.  If it is possible to enter interp_only_mode we add
2309   // the code to check if the event should be sent.
2310   if (JvmtiExport::can_post_interpreter_events()) {
2311     Label L;
2312     Register temp_reg = O5;
2313     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2314     ld(interp_only, temp_reg);
2315     cmp_and_br_short(temp_reg, 0, equal, pt, L);
2316     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2317     bind(L);
2318   }
2319 
2320   {
2321     Register temp_reg = O5;
2322     SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2323     call_VM_leaf(noreg,
2324       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2325       G2_thread, Lmethod);
2326   }
2327 
2328   // RedefineClasses() tracing support for obsolete method entry
2329   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2330     call_VM_leaf(noreg,
2331       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2332       G2_thread, Lmethod);
2333   }
2334 }
2335 
2336 
2337 // Inline assembly for:
2338 //
2339 // if (thread is in interp_only_mode) {
2340 //   // save result
2341 //   InterpreterRuntime::post_method_exit();
2342 //   // restore result
2343 // }
2344 // if (DTraceMethodProbes) {
2345 //   SharedRuntime::dtrace_method_exit(thread, method);
2346 // }
2347 //
2348 // Native methods have their result stored in d_tmp and l_tmp
2349 // Java methods have their result stored in the expression stack
2350 
2351 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method,
2352                                                    TosState state,
2353                                                    NotifyMethodExitMode mode) {
2354   // C++ interpreter only uses this for native methods.
2355 
2356   // Whenever JVMTI puts a thread in interp_only_mode, method
2357   // entry/exit events are sent for that thread to track stack
2358   // depth.  If it is possible to enter interp_only_mode we add
2359   // the code to check if the event should be sent.
2360   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2361     Label L;
2362     Register temp_reg = O5;
2363     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2364     ld(interp_only, temp_reg);
2365     cmp_and_br_short(temp_reg, 0, equal, pt, L);
2366 
2367     // Note: frame::interpreter_frame_result has a dependency on how the
2368     // method result is saved across the call to post_method_exit. For
2369     // native methods it assumes the result registers are saved to
2370     // l_scratch and d_scratch. If this changes then the interpreter_frame_result
2371     // implementation will need to be updated too.
2372 
2373     save_return_value(state, is_native_method);
2374     call_VM(noreg,
2375             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2376     restore_return_value(state, is_native_method);
2377     bind(L);
2378   }
2379 
2380   {
2381     Register temp_reg = O5;
2382     // Dtrace notification
2383     SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2384     save_return_value(state, is_native_method);
2385     call_VM_leaf(
2386       noreg,
2387       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2388       G2_thread, Lmethod);
2389     restore_return_value(state, is_native_method);
2390   }
2391 }
2392 
2393 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) {
2394 #ifdef CC_INTERP
2395   // result potentially in O0/O1: save it across calls
2396   stf(FloatRegisterImpl::D, F0, STATE(_native_fresult));
2397 #ifdef _LP64
2398   stx(O0, STATE(_native_lresult));
2399 #else
2400   std(O0, STATE(_native_lresult));
2401 #endif
2402 #else // CC_INTERP
2403   if (is_native_call) {
2404     stf(FloatRegisterImpl::D, F0, d_tmp);
2405 #ifdef _LP64
2406     stx(O0, l_tmp);
2407 #else
2408     std(O0, l_tmp);
2409 #endif
2410   } else {
2411     push(state);
2412   }
2413 #endif // CC_INTERP
2414 }
2415 
2416 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) {
2417 #ifdef CC_INTERP
2418   ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0);
2419 #ifdef _LP64
2420   ldx(STATE(_native_lresult), O0);
2421 #else
2422   ldd(STATE(_native_lresult), O0);
2423 #endif
2424 #else // CC_INTERP
2425   if (is_native_call) {
2426     ldf(FloatRegisterImpl::D, d_tmp, F0);
2427 #ifdef _LP64
2428     ldx(l_tmp, O0);
2429 #else
2430     ldd(l_tmp, O0);
2431 #endif
2432   } else {
2433     pop(state);
2434   }
2435 #endif // CC_INTERP
2436 }
2437 
2438 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2439 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2440                                                         int increment, int mask,
2441                                                         Register scratch1, Register scratch2,
2442                                                         Condition cond, Label *where) {
2443   ld(counter_addr, scratch1);
2444   add(scratch1, increment, scratch1);
2445   if (is_simm13(mask)) {
2446     andcc(scratch1, mask, G0);
2447   } else {
2448     set(mask, scratch2);
2449     andcc(scratch1, scratch2,  G0);
2450   }
2451   br(cond, false, Assembler::pn, *where);
2452   delayed()->st(scratch1, counter_addr);
2453 }